INDIAN INSTITUTE OF TECHNOLOGY DELHI

Term Paper
“Unit commitment Through Evolutionary
Techniques”

Submitted To: Submitted By:

Dr. P R Bijwe ABHIJEET DHANDA

EES123008

M. Tech (Power System)

Department of Electrical Engg.

 Unit Commitment

ABSTRUCT: Unit Commitment is one of the most important problems of the power system.
Unit Commitment (UC) means scheduling of available power plants for generation by taking
into account, technological aspects (such as minimal operation point, minimum up and down
time and transient behavior), economical considerations (such as start-up costs and operational
costs), social elements (such as availability of staff and work-schemes) and environmental
aspects (Minimum Emission). UC is highly non linear, high dimensional and mixed-integer
combinatorial optimization problem with both binary/integer (unit status variable) and
continuous (unit output power) variables. In addition to this, the demand and reserve requirement
and constraints of the generating units is to be satisfied. This makes UC largest problem in
Power system.
Unit Commitment problem has been solved by different optimization methods like priority list,
dynamic programming, Lagrangian relaxation, Genetic algorithms and evolutionary methods. In
this paper various evolutionary programming methods for UC problem is discussed.

A Solution to the Unit Commitment Problem Using Imperialistic Competition Algorithm

Moosa Moghimi Hadji, Student Member, IEEE, and Behrooz Vahidi, Senior Member, IEEE

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 27, NO. 1, FEBRUARY 2012

In this paper a new evolutionary algorithm based on Imperialistic competition is used for UC
problem. In earlier days of Imperialism in world all the countries are divided into imperialists
and colonies. The colonies are divided among imperialists. The imperialists and colonies all
together form some empires. The imperialistic competition means fighting between the empires
to take hold of colonies. At the end of competition only one empire remains with colonies and
imperialist have same cost and thus the colonies of world converge to the global optimum of the
objective function.
In Unit commitment the main objective is the minimum operating cost of generating units for a
particular operating period with all the system and unit constraint satisfied. The Fuel Cost
Function of the kth unit is calculated as
CFk(p) = ak + bk p + ck p2 where ak, bk, ck are the fuel cost coefficient of the kth generator.
Operating Cost includes fuel, start up and shut down costs. In this paper start up costs are
expressed as linear and exponential function of time and only the fuel cost for a time horizon and
start up cost are considered. Various constraints like Min. and Max. power limits, Ramp-up and
down rate, Power balance, Spinning reserve, Min up-down times are considered for formulation
of UC problem.
Concept of Imperialistic Competition:
Like other Evolutionary algorithms, it start with some initial population of countries, the
countries having best value of objective function is considered as imperialistic countries and all
other will make a colonies. The division of colonies among the imperialistic country is based on
the power possessed by it. More power means less cost in this algorithm. Thus for cost
minimization problem is directly related to power maximization problem. Thus ICA consists of
following steps:
1. Generating initial empires: From an initial population of countries, the cost function of a
particular country is evaluated depend upon the position of the country in N dimensional
space. Depend upon the cost function of the country; it is classified as Imperialistic and
colonies. Thus more cost means less power in this algorithm. In order to divide the
countries among the imperialistic and form the empires. The initial no. of colonies
possessed by the empires depends upon the power of the empire. A normalized cost
function is defined and normalized power of each imperialistic country is defined and the
no. of colonies of the empire is the product of normalized power and total no of colonies.
2. Moving colonies towards the empires: In this stage, the colonies start to move toward
their relevant imperialists. The positions of the colonies of the empire are updated. It is
like a PSO problem where position and velocity are updated at every iteration.
3. Exchanging the positions: During the previous stage, a colony may reach to a position
with lower cost than that of the imperialist. In such a case, the positions of the imperialist
and that colony must be exchanged. Then the rest of the colonies of this empire move
toward the new position of the imperialist
4. Calculate the total power of the empire: The total power of an empire depends on both
the power of the imperialist and the power of its colonies.
5. Imperialistic competition: In Imperialistic competition, all the empires try to take
possession of colonies of other empire. By choosing the weakest colony of the weakest
empire and calculating the possession probability of each empire and forming a vector
 consists of PS (PoSsession probability) of all the empires. A new random vector R having
the same size as PS is created and difference of these vectors will give the vector D
which gives the weakest colony to that empire having largest value of element in D
vector.
6. Eliminating Empire when all the colonies are lost by the empire and converge when all
the empire except the most powerful remain and all the colonies of that have the same
cost as the imperialist in the empire.
Implementation of ICA for unit commitment:

The position of the country is defined by the sequence of ON/OFF cycles of a particular unit
with positive value considered for duration of ON time and negative value for the duration of
OFF time. In this paper the no. of ON/OFF cycles for peak load is considered 5. Thus the size of
vector of a country is 5 times no. of generating units. For generating the initial population the
units are divided into three classes as base load, medium load and peak loads. In this paper only
the optimal state of medium load is to be considered as the base load are fixed and peak loads are
depend upon the fuel cost. Priority list is created for the medium loads and a random permutation
of medium loads are placed between the peak and base loads

For the cost function is sum of all the production costs over an time horizon and penalty
functions for violation of constraints. In this paper two constraints are used, one for spinning
reserve constraint violations and other for excessive capacity for situations in which the loads are
less than sum of all the committed units. Thus the cost function becomes:

CF= TC + ∏ (res) + ∏ (exc)

The constraint handling like minimum up and down time constraints are presented in this paper.
In this the duration of first cycle of each unit initialized as T 1i which is the first schedule for ith
unit. It is then compared with initial state IS i . The initial value is calculates as it checks whether
the last cycle of the previous day schedule and first cycle of present schedule must also satisfy
the minimum up and down time. For example if last cycle of previous day schedule is 5 hrs and
minimum up time 7 hrs then the first cycle should be at least 2 hrs and it also checks the
minimum down time also if the first cycle is shut down schedule. In the next cycle, a new
variable is added to check the sum of absolute values of cycles of the unit is equal to the
scheduling horizon. For each cycles, residue is the difference between the scheduling period and
sum of absolute values of previous cycle of unit.

The cost function is then calculated after satisfying all the time constrains and then the
imperialistic algorithm is used for obtaining the optimum.

Benefits: The use of Integer coded method helps in incorporating the minimum up and down
time constraints in cost function itself thus avoiding the use any other penalty function for these
constraints.

Results shows that increasing the no. of countries will improves the convergence of solution and
mean operating cost but computational time increases. This method when compared with Integer
coded GA method shows the increase in time but quality of solution is much better than ICGA.

Bacterial Foraging-Based Solution to the Unit-Commitment Problem

Morteza Eslamian, Seyed Hossein Hosseinian, and Behrooz Vahidi, Senior Member, IEEE

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 24, NO. 3, AUGUST 2009

This paper employs a new evolutionary algorithm known as bacterial foraging (BF) for solving
the UC problem. This new integer-code algorithm is on the base of foraging behavior of E-coli
Bacteria in the human intestine. By integer coding of the problem, computation time decreases
and the minimum up/down-time constraints may be coded directly and thus there is no need for
any other penalty function. Thus this method is also integer coded method as ICA method shown
in first paper.
Concept of Bacterial forging optimization:
This optimization depends upon the forging strategies used by organisms and selection of good
from bad methods of forging. Thus forging strategies is governed by four processes:
1. Chemotaxis: This means movement of the organism in favorable direction. If the next
position is more favorable/reduce cost then movement will occur in the same direction.
2. Swarming: This means attracting other bacteria makes groups so that they can reach the
desired place more rapidly.
3. Reproduction: least healthy bacteria will die and healthy bacteria will spilt into two to
make the population size same.
 4. Elimination and dispersal: Due to some event there is possibility of change in the life of
population. These events can kill or disperse bacteria. Thus by dispersing there is
possibility of obtaining the better positions and prevent bacteria from trapped in local
minima.
In this paper improved bacterial forging optimization algorithm is shown in which in the
Chemotaxis stage the minimum value of all the Chemotaxis cost is used to evaluate the bacteria
health and during reproduction stage after choosing the healthy bacteria it is first placed into the
minimum cost location from Chemotaxis stage and then it is split into two bacteria for making
population constant. Various loops like elimination and dispersal loop, reproduction loop &
Chemotaxis loop start the iterative process.
Implementation of BFA for unit commitment:
Total production cost of unit consists of fuel costs, start-up costs and shutdown costs. Fuel costs
are calculated using heat rate and fuel price information. Start-up costs are expressed as an
exponential or linear function of the time. Shut down costs are considered to be fixed cost.
Various constraints like power output limits up/down limits, power balance, ramp up and ramp
down limit and spinning reserve limit is considered.
The location of bacteria (θ) is defined by sequence of integer numbers, representing the sequence
of ON/OFF cycle durations of units during the unit commitment horizon. This makes the size of
θ to be reduced. In this case the last cycle is not considered as independent parameter as it
depends upon the total time schedule. Thus last cycle is the difference of total time – sum of
cycle durations of C-1 cycle (consider C to be total no. of cylces). Positive value means ON
duration and negative value means OFF duration. Initial population is calculated in same process
as in ICA method with initialization of the first cycle of unit which depends upon the minimum
UP/DOWM time constraints of last cycle of previous schedule. Fitness function is evaluated
depend upon the cost function and penalty function to penalize for violation of system
constrains.
In chemotactic process, each bacterium moves in direction of random vector with step size to
reach better position. Thus the movement vector is chosen such that the value should be integer
type as in UC parameters are in integer value. Also value should be so chosen that the Economic
dispatch calculation is to be done only for the changed generation which is obtained by adding
the previous schedules and movement vector. This saves a lot of computational time saving as
only for one or two particular units economic dispatch is to be calculated. Time constraints are
implicit used in the initial selection of population of bacteria itself and economic dispatch is
performed only for the selected units without any time constraints.
Two types of converge criteria is used, first without swarming and then with swarming is used
and observed better converge with swarming included.
This paper also compares the BF methods (1) which use penalty functions for time constraints
and BF method (2) which uses integer coded of problem and thus no need for penalty function.
On comparison method (2) solves the problem with no penalty violation and less time as
compared to method (1). Also this method (2) is compared with LR and GA method and gives
better results for large no. of unit system.

Multi-objective unit commitment problem with reliability function using fuzzified binary
real coded artificial bee colony algorithm
K. Chandrasekaran S.P. Simon

IET Generation, Transmission & Distribution , 13th June , 2011

This paper proposes a method that employs fuzzified binary real coded artificial bee colony
(ABC) algorithm for solving multi-objective unit commitment problem. On solving unit
commitment problem, the proposed binary coded ABC algorithm finds the ON/OFF status of the
generating units whereas the economic dispatch is solved using the real coded ABC. Here, three
conflicting functions such as fuel cost, emission and reliability level of the system are
considered. Fuzzy set theory is used to formulate the functions into single objective problem
This method uses new constraints like emission limitation and a separate objective function of
reliability maximization to increase the reliability. Fuzzy set theory helps in transforming multi
objective function into single objective function.

Concept of Artificial bee colony algorithm:

The artificial bee colony (ABC) algorithm is a swarm-based meta-heuristic algorithm for
optimising numerical problems. The model consists of four essential components: employed
bees, onlooker bees or unemployed bees, food sources and dancing area. Employed bees search
for food depending on their own experience. Once the employed bees complete their search
process, information sharing is done by dancing. The exact distance and quality of food is known
by the angle α between the Sun’s position and direction of food source and time of dance.
Now, the onlookers produce a modification in the position selected by it using (1) and evaluate
the nectar amount of the new source.
V pq = x pq + θ pq ( x pq- x fq) (1)
Where p, q & f are randomly chosen indexes
If the nectar amount of the new source is higher than that of the previous one, the onlookers
remember the new position; otherwise, it retains the old one.
In other words, a greedy selection method is employed as the selection operation between old
and new food sources.
In MOUCP, binary numbers 1 and 0 are used to indicate the unit status ON/OFF. In this paper,
the relevant variables are interpreted in terms of changes of probability.
The onlookers produce a modification in the position and evaluate the nectar amount of V pqthe
new source. Thus in (1) if value of V pq takes the value of [-1, 2] if the value of θ pq lies between [-
1, 1]. Thus the value is compared with some threshold limit. If V pq is greater than threshold then
1 is chosen otherwise 0 is chosen.

Implementation of BRABC for unit commitment :

1. Initialize the parameters like colony size, employed and onlookers bees etc
2. Initial generation of population in binary coded ABC is done by randomly generating a
population of M solutions by binary string
3. Check for any time constraint/ spinning reserve constraint and if there is violation then
unit status is to be changed to maintain the constrain. This is called repair scheme.
4. Fitness function (FF) is evaluated for each population and set the count as 1
5. Modification of the position as per the equation (1) and again check for any time/spinning
reserve constraints and perform repair scheme if any volition occur.
6. Economic dispatch is done and FF is evaluated for each modified position. It is then
compared with FF (old) and If it is better than FF (old) the Limiting value is set.
7. Recruit onlookers bees which choose the food source depend upon the information from
employed bees.
8. Modification of position of onlookers bees as point no. 5. Check for any limit voliations
9. Memories the best solution so far.
10. Stop the process if termination criteria is satisfied.
This paper shows good convergence behavior if the control parameter colony size is optimally
tuned. The results shows the superiority of this method over others in solving multi objective
problem.
Optimal Deviation Based Firefly Algorithm Tuned Fuzzy Design for Multi-Objective UCP

K. Chandrasekaran, StudentMember, IEEE, and Sishaj P. Simon

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 1, FEBRUARY 2013

In this paper, a multi objective Unit commitment problem consists for relaibilty requirement is
solved with optimal deviation based firefly algorithm tuned fuzzy membership function. The
ON/OFF status of the generating units is obtained by binary coded FF whereas the sub-problem
economic dispatch (ED) is obtained by real coded FF. Here the conflicting functions are
formulated as a single objective function using fuzzy weighted optimal deviation. This paper is
inspired by the flashing behavior of the fireflies, and compared its performance on benchmark
functions with other heuristic optimization algorithms. In this paper spinning reserve constraints
are eliminated and new reliability function is incorporated as a objective function thus making ita
multi objective function.
The MOUCP is formulated as min ( Fc, LOLP) where LOLP is loss of load probability which is
related to reliability of the system. To convert it into single objective function calculate the best
values of fuel cost and LOLP separately considering only one objective function. The percentage
deviation from these best values is calculated and the multi objective function is converted to
single objective function
MOF = σ cost + σ LOLP where σ cost& σ LOLP is the percentage deviation from the best values.
By giving the fuzzy weights to this function it is converted into the FMOF
FMOF= μcost σ cost + μ LOLP σ LOLP
Fitness function is then calculated as
FIT = 1/ (1+ FMOF)
Concept of Fire Fly algorithm:
Firefly algorithm is a novel nature-inspired algorithm inspired by social behavior of fireflies. It
uses flashing characteristics of fireflies for optimization of solution. It uses the concept of
attracting of flies to each other with less brighter one move towards the brighter one and more
brightness means less distance between the flies. If the brightness is same then they will move
randomly. Finally the brightness of a firefly is determined by the value of the objective function.
In the FF algorithm, the main form of attractiveness function β can be any of the monotonically
decreasing functions as given below:

The modification of the firefly position is determined by

Where fly p is attracted to another more brighter fly q. The second term is due to attraction and
third term introduces randomization with α being the randomized parameter.
Implementation of BRCFF for unit commitment:

1. Initialize the parameters like firefly position etc

2. Initial generation of population in binary coded FF is done by randomly generating a
population of M solutions or firefly positions by binary string
3. Check for any time constraint/ spinning reserve constraint and if there is violation then
unit status is to be changed to maintain the constrain. This is called repair scheme.
All other steps are nearly the same as in the previous paper and thus require the real coded value
for the determination of the fitness function value and thus require separate loop for the real
coded FF values.
The BRCFF, when applied to practical UCP with deterministic and probabilistic reserve
constraints, gives the minimum operating cost solution when compared with other conventional
and non-conventional techniques reported in the literature

A New Quantum-Inspired Binary PSO: Application to Unit Commitment Problems for

Power Systems
Yun-Won Jeong, Jong-Bae Park, Member, IEEE, Se-Hwan Jang, and Kwang Y. Lee, Fellow, IEEE
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 25, NO. 3, AUGUST 2010

This paper proposes a new binary particle swarm optimization (BPSO) approach inspired by
quantum computing, namely quantum-inspired BPSO (QBPSO). The proposed QBPSO
combines the conventional BPSO with the concept and principles of quantum computing such as
a quantum bit and superposition of states.
The QBPSO adopts a Q-bit individual for the probabilistic representation, which replaces the
velocity update procedure in the particle swarm optimization. The proposed QBPSO is applied to
unit commitment (UC) problems for power systems which are composed of up to 100-units with
24-h demand horizon.
The QBPSO introduces a Q-bit individual for the probabilistic representation of a particle,
thereby replacing the velocity update procedure in the traditional PSO. In QBPSO, therefore, an
inertia weight factor and two acceleration coefficients can be removed and only one factor,
rotation angle, is needed when modifying the position of particles.
In Binary PSO, position vectors of a particle are binary one. Thus each particle have large no of
elements (1 or 0s). In this another function called probability threshold is defined as sigmoid
function. Thus if the random no. selected from the set [0,1] is less than sigmoid value then new
position will be 1 otherwise 0.
Concept of Quantum inspired PSO algorithm:
The smallest unit of information stored in a quantum computer is called a quantum bit or Q-bit.
A Q-bit is analogous to a bit of storage in a traditional computer. A Q-bit may be in the “1” state,
in the “0” state, or in any superposition of the two, while a bit in traditional computing can only
hold a single state, either 0 or 1. To illustrate this, the traditional 0 and 1 values are
written as and , and the state of a Q-bit is represented as follows:
ψ = α 0 > + β 1>
where α and β are complex numbers that specify the probability amplitudes of the
corresponding states. The state of a Q-bit can be changed by the operation with a quantum gate
such as NOT gate, rotation gate, and Hadamard gate, etc. Q bit individual is updated through a
rotation gate operator
In the proposed QBPSO, the state of each element in a particle takes a value of 0 or 1 by the
probability α2 or β2. Thus the velocity update process is replaced by the quantum computing.
Implementation of QBPSO for unit commitment:
The objective is to minimize the total operating cost, fuel cost, start up and shut down cost of all
units for a time horizon subject to system and unit constraints. Since the UC problems involve
determining the on/off states of generating units, the decision variables are the on/off status of
generating units. The process is as described :
1. Creating initial Q bit Indiviual and position of particles
2. Q bit update by the rotation gate
3. Modification of position of particles
4. Update of Pbest and G best values
 5. Stopping creitra
The results of this algorithm is compared with convential BPSO and Improved quantum
evolutionary algorithm (IQEA) with the same contraints. The results shows better solutions with
QBPSO method. A comparsion with other method like EA, GA, SA methods also shows better
solutions quality with the QBPSO method.